Suppress some warnings in WebGPU EP generated by GCC 13 (#23984)

### Description

Replace #23445, resolve conflicts and add one new file.

---------

Co-authored-by: Changming Sun <chasun@microsoft.com>

Author: fs-eire

onnxruntime/core/providers/webgpu/buffer_manager.h

@@ -5,7 +5,7 @@
 
 #include <iosfwd>
 
-#include <webgpu/webgpu_cpp.h>
+#include "core/providers/webgpu/webgpu_external_header.h"
 
 #include "core/framework/execution_provider.h"
 

onnxruntime/core/providers/webgpu/compute_context.h

@@ -3,7 +3,7 @@
 
 #pragma once
 
-#include <webgpu/webgpu_cpp.h>
+#include "core/providers/webgpu/webgpu_external_header.h"
 
 #include <utility>
 

onnxruntime/core/providers/webgpu/program_manager.h

@@ -6,7 +6,7 @@
 #include <string>
 #include <unordered_map>
 
-#include <webgpu/webgpu_cpp.h>
+#include "core/providers/webgpu/webgpu_external_header.h"
 
 #include "core/common/common.h"
 

onnxruntime/core/providers/webgpu/shader_helper.h

@@ -5,7 +5,7 @@
 
 #include <sstream>
 
-#include <webgpu/webgpu_cpp.h>
+#include "core/providers/webgpu/webgpu_external_header.h"
 
 #include "core/framework/tensor_shape.h"
 

onnxruntime/core/providers/webgpu/tensor/pad.cc

@@ -146,27 +146,21 @@ Status Pad::ComputeInternal(ComputeContext& context) const {
       uint16_t value = math::floatToHalf(value_);
       std::memcpy(&value_uint32, &value, sizeof(value));
     } else {
-      value_uint32 = *reinterpret_cast<const uint32_t*>(&value_);
+      std::memcpy(&value_uint32, &value_, sizeof(value_uint32));
     }
   } else if (value_tensor) {
     ORT_ENFORCE(value_tensor->DataType() == input_tensor->DataType() && value_tensor->Shape().Size() == 1,
                 "Value tensor should be a 1D tensor of size 1 with the same type as that of the input tensor");
     switch (data_type) {
-      case ONNX_NAMESPACE::TensorProto_DataType_INT32: {
-        int32_t value = value_tensor->Data<int32_t>()[0];
-        value_uint32 = *reinterpret_cast<uint32_t*>(&value);
-      } break;
-      case ONNX_NAMESPACE::TensorProto_DataType_FLOAT: {
-        float value = value_tensor->Data<float>()[0];
-        value_uint32 = *reinterpret_cast<uint32_t*>(&value);
-      } break;
       case ONNX_NAMESPACE::TensorProto_DataType_FLOAT16: {
         uint16_t value = value_tensor->Data<MLFloat16>()[0].val;
         std::memcpy(&value_uint32, &value, sizeof(value));
       } break;
-      case ONNX_NAMESPACE::TensorProto_DataType_UINT32: {
-        value_uint32 = value_tensor->Data<uint32_t>()[0];
-      } break;
+      case ONNX_NAMESPACE::TensorProto_DataType_INT32:
+      case ONNX_NAMESPACE::TensorProto_DataType_FLOAT:
+      case ONNX_NAMESPACE::TensorProto_DataType_UINT32:
+        std::memcpy(&value_uint32, value_tensor->DataRaw(), sizeof(value_uint32));
+        break;
       default:
         return ORT_MAKE_STATUS(ONNXRUNTIME, INVALID_ARGUMENT, "Unsupported input type: ", static_cast<int>(data_type));
     }

onnxruntime/core/providers/webgpu/webgpu_context.h

@@ -6,7 +6,7 @@
 #include <memory>
 #include <mutex>
 
-#include <webgpu/webgpu_cpp.h>
+#include "core/providers/webgpu/webgpu_external_header.h"
 
 #include "core/common/common.h"
 #include "core/framework/library_handles.h"

onnxruntime/core/providers/webgpu/webgpu_pix_frame_generator.h

@@ -14,7 +14,7 @@
 
 #include <memory>
 
-#include <webgpu/webgpu_cpp.h>
+#include "core/providers/webgpu/webgpu_external_header.h"
 
 namespace onnxruntime {